In [ ]:
import numpy as np
import cv2
import glob
import matplotlib.pyplot as plt
%matplotlib qt
# prepare object points, like (0,0,0), (1,0,0), (2,0,0) ....,(6,5,0)
objp = np.zeros((9*6,3), np.float32)
objp[:,:2] = np.mgrid[0:9, 0:6].T.reshape(-1,2)
# Arrays to store object points and image points from all the images.
objpoints = [] # 3d points in real world space
imgpoints = [] # 2d points in image plane.
# Make a list of calibration images
images = glob.glob('camera_cal/calibration*.jpg')
# Step through the list and search for chessboard corners
for idx, fname in enumerate(images):
img = cv2.imread(fname)
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
# Find the chessboard corners
ret, corners = cv2.findChessboardCorners(gray, (9,6), None)
# If found, add object points, image points
if ret == True:
objpoints.append(objp)
imgpoints.append(corners)
# Draw and display the corners
cv2.drawChessboardCorners(img, (9,6), corners, ret)
#write_name = 'corners_found'+str(idx)+'.jpg'
#cv2.imwrite(write_name, img)
cv2.imshow('img', img)
cv2.waitKey(500)
cv2.destroyAllWindows()
If the above cell ran sucessfully, you should now have objpoints and imgpoints needed for camera calibration. Run the cell below to calibrate, calculate distortion coefficients, and test undistortion on an image¶
In [ ]:
import pickle
%matplotlib inline
# Test undistortion on an image
img = cv2.imread('camera_cal/calibration1.jpg')
img_size = (img.shape[1], img.shape[0])
# Do camera calibration given object points and image points
ret, mtx, dist, rvecs, tvecs = cv2.calibrateCamera(objpoints, imgpoints, img_size,None,None)
print (dist)
dst = cv2.undistort(img, mtx, dist, None, mtx)
cv2.imwrite('reportPict/undist_calibration1.jpg',dst)
# Save the camera calibration result for later use (we won't worry about rvecs / tvecs)
dist_pickle = {}
dist_pickle["mtx"] = mtx
dist_pickle["dist"] = dist
pickle.dump( dist_pickle, open( "camera_cal/wide_dist_pickle.p", "wb" ) )
#dst = cv2.cvtColor(dst, cv2.COLOR_BGR2RGB)
# Visualize undistortion
f, (ax1, ax2) = plt.subplots(1, 2, figsize=(20,10))
ax1.imshow(img)
ax1.set_title('Original Image', fontsize=30)
ax2.imshow(dst)
ax2.set_title('Undistorted Image', fontsize=30)
In [1]:
import numpy as np
import cv2
import glob
import pickle
import matplotlib.pyplot as plt
from moviepy.editor import VideoFileClip
from IPython.display import HTML
%matplotlib qt
%matplotlib inline
def convertToBinMix(img, yellowH=(20, 30), yellowS=(40, 255), whiteL=(205,255), sxThresh=(10, 255), lxThresh=(40, 255), dirThreshold=(0.8, 1.1)):
"""
This functions returns binary image where lines are marked with '1'
Parameters
----------
img : array
Input BGR image
yellowH : tuple
threshold for yellow Hue channel
yellowS : tuple
threshold for yellow Saturation channel
whiteL : tuple
threshold for white light channel
sxThresh : tuple
threshold for X sobel over saturation channel
sxThresh : tuple
threshold for X sobel over light channel
dirThreshold
threshold for sobel direction over light and saturation channel
Returns
-------
float
Calculated radius
"""
img = np.copy(img)
# Convert to HLS color space and separate the V channel
hls = cv2.cvtColor(img, cv2.COLOR_BGR2HLS).astype(np.float)
hChannel = hls[:,:,0]
lChannel = hls[:,:,1]
sChannel = hls[:,:,2]
# Sobel x over l channel
sobelx = cv2.Sobel(lChannel, cv2.CV_64F, 1, 0) # Take the derivative in x
sobely = cv2.Sobel(lChannel, cv2.CV_64F, 0, 1) # Take the derivative in y
direction = np.arctan2(np.absolute(sobely), np.absolute(sobelx))# Calculate the direction of the gradient in radians
absSobelx = np.absolute(sobelx) # Absolute x derivative to accentuate lines away from horizontal
scaledSobel = np.uint8(255*absSobelx/np.max(absSobelx))
# Threshold x gradient
lxBinary = np.zeros_like(scaledSobel)
lxBinary[(scaledSobel >= lxThresh[0]) & (scaledSobel <= lxThresh[1]) & (direction >= dirThreshold[0]) & (direction <= dirThreshold[1])] = 1
# Sobel x over s channel
sobelx = cv2.Sobel(sChannel, cv2.CV_64F, 1, 0) # Take the derivative in x
sobely = cv2.Sobel(sChannel, cv2.CV_64F, 0, 1) # Take the derivative in y
direction = np.arctan2(np.absolute(sobely), np.absolute(sobelx))# Calculate the direction of the gradient in radians
absSobelx = np.absolute(sobelx) # Absolute x derivative to accentuate lines away from horizontal
scaledSobel = np.uint8(255*absSobelx/np.max(absSobelx))
# Threshold x gradient
sxBinary = np.zeros_like(scaledSobel)
sxBinary[(scaledSobel >= sxThresh[0]) & (scaledSobel <= sxThresh[1]) & (direction >= dirThreshold[0]) & (direction <= dirThreshold[1])] = 1
# Threshold yellow pixels
yellowBinary = np.zeros_like(hChannel)
yellowBinary[(hChannel >= yellowH[0]) & (hChannel <= yellowH[1]) & (sChannel >= yellowS[0]) & (sChannel <= yellowS[1])] = 1
# threshold white pixels
whiteBinary = np.zeros_like(lChannel)
whiteBinary[(lChannel >= whiteL[0]) & (lChannel <= whiteL[1])] = 1
# Stack each channel
resultBin = np.zeros_like(whiteBinary)
resultBin[(yellowBinary == 1) | (whiteBinary == 1) | (lxBinary == 1) | (sxBinary == 1)] = 1
return resultBin
def convertToBinColors(image):
hls = cv2.cvtColor(image, cv2.COLOR_BGR2HLS).astype(np.float)
#bgr thresholding for yellow
lower = np.array([0,140,170],dtype = "uint8")
upper = np.array([170, 255, 255],dtype = "uint8")
bgrYellow = cv2.inRange(image, lower, upper)
#bgr thresholding for white
lower = np.array([200,100,100],dtype = "uint8")
upper = np.array([255, 255, 255],dtype = "uint8")
bgrWhite = cv2.inRange(image, lower, upper)
#hls thresholding for yellow
lower = np.array([15,90,40],dtype = "uint8")
upper = np.array([30, 255, 255],dtype = "uint8")
hlsYellow = cv2.inRange(hls, lower, upper)
binary = np.zeros_like(hlsYellow)
binary [(hlsYellow == 255) | (bgrYellow==255) | (bgrWhite==255)]= 1
return binary
def convertToBinNotGray(image):
#difference between chnnels is small for gray
b = image[:,:,0].astype('int16')
g = image[:,:,1].astype('int16')
r = image[:,:,2].astype('int16')
bMinG = abs(b-g).astype('uint8')
bMinR = abs(b-r).astype('uint8')
gMinR = abs(g-r).astype('uint8')
diff=20
diffGray = np.zeros_like(bMinG)
#diffGray[(bMinG <= diff) & (bMinR <= diff) & (gMinR <= diff)] = 1 #challenge
diffGray[(bMinG <= diff) & (gMinR <= diff)] = 1
#bgr thresholding for gray
lower = np.array([0,0,0],dtype = "uint8")
upper = np.array([185, 185, 185],dtype = "uint8")
maybeBgrGray = cv2.inRange(image, lower, upper)
upper = np.array([120, 120, 120],dtype = "uint8")
alwaysBgrGray = cv2.inRange(image, lower, upper)
#hls thresholding for gray
hls = cv2.cvtColor(image, cv2.COLOR_BGR2HLS)
lower = np.array([8,0,0],dtype = "uint8")
upper = np.array([18, 200, 90],dtype = "uint8")
hlsGray = cv2.inRange(hls, lower, upper)
binary = np.zeros_like(diffGray)
binary [((maybeBgrGray == 0) | (diffGray==0)) & (alwaysBgrGray == 0) & (hlsGray==0)]= 1
#binary [((maybeBgrGray == 0) | (diffGray==0)) & (alwaysBgrGray == 0)]= 1
#binary [diffGray == 0]= 1
return binary
ymPerPix = 3.0/(358.0-462.0) # based on dashed line measured on bird eye view and real life line legth of 3m
xmPerPix = 3.7/(978.0-302.0) # based on line width measured on bird eye view and real life line width of 3.7m
# polyCoef - Coeficients of polymial of left line in pixel space
# yEval - where radius should be calculated in y pixels
def calculateRadius(polyCoef, yEval):
"""
calculate radius of quadratic function
Parameters
----------
polyCoef : array
Coeeficients of quadratic poly fit
yEval : float
Argument for which radius is calculated
Returns
-------
float
Calculated radius
"""
#x= mx / (my ** 2) *a*(y**2)+(mx/my)*b*y+c
polyCoefMeters = np.array([polyCoef[0]*(xmPerPix/(ymPerPix**2)), polyCoef[1]*(xmPerPix/ymPerPix), polyCoef[2]])
# Calculate the radius of curvature
#Rcurve=((1+(2Ay+B)^2)^3/2)/∣2A∣
radius = ((1 + (2*polyCoefMeters[0]*yEval*ymPerPix + polyCoefMeters[1])**2)**1.5) / np.absolute(2*polyCoefMeters[0])
return radius
def squarefit(pixX, pixY):
"""
Fit a second order polynomial
Parameters
----------
pixX : array
X coordinates of pixels to be fitted
pixY : array
Y coordinates of pixels to be fitted
Returns
-------
array
Coeeficients of quadratic poly fit
"""
if pixX.size == 0 or pixY.size == 0:
return None
# Fit a second order polynomial
return np.polyfit(pixY, pixX, 2)
def calcQuadFitVal(squareFit, arg):
"""
calculate quadratic function value
Parameters
----------
squareFit : array
Coeeficients of quadratic poly fit
arg : float
Argument for which value is calculated
Returns
-------
float
Value for given argument
"""
return squareFit[0]*arg**2 + squareFit[1]*arg + squareFit[2]
def pixelsSanityCheck(pixY, minpixels, imgShape):
"""
Perform sanity checks on line pixels
Parameters
----------
pixY : array
Y coordinates of line pixels
minpixels : int
Minimum number of pixels
imgShape: array
shape of image
Returns
-------
bool
True if check was successful False otherwise
"""
middleY = int(imgShape[0]/2)
if len(pixY) < minpixels or np.amax(pixY) < middleY or np.amin(pixY) > middleY:
return False
return True
def lineFitSanitycheck(leftFit, rightFit, margin, imgShape):
"""
Perform sanity checks on fitted lines
Parameters
----------
leftFit : array
Coeeficients of left line quadratic poly fit
rightFit : array
Coeeficients of right line quadratic poly fit
margin : integer
margin detected line width
imgShape: array
shape of image
Returns
-------
bool
True if check was successful False otherwise
"""
bottomY = imgShape[0]
imageHalf = int(imgShape[1]/2)
bottomRight = calcQuadFitVal(rightFit, bottomY)
bottomLeft = calcQuadFitVal(leftFit, bottomY)
#check if lines are on correct sides
if bottomLeft > imageHalf or bottomRight < imageHalf:
return False
bottomWidth = bottomRight - bottomLeft
#check if line is wide enougth
if bottomWidth < (imageHalf - margin):
return False
#check if line width does not warry more than margin
for y in range(bottomY-1):
widthDiff = abs((calcQuadFitVal(rightFit, y) - calcQuadFitVal(leftFit, y)) - bottomWidth)
if widthDiff > margin:
return False
return True
def markLines(img, undistImg, prevLeftFit = None, prevRightFit = None, margin = 100, minpix = 50, nwindows = 9):
"""
This functions detect lines, calculate curvature and car position using binary bird eye image and marks them on BGR image
Parameters
----------
img : array
input bird eye view binary image
undistImg : array
input undistorted image on which lines and curvatures will be marked
prevLeftFit : array
Coeeficients of lefy line quadratic poly fit in previous frame
prevRightFit : array
Coeeficients of right line quadratic poly fit in previous frame
margin : integer
search margin for new line
minpix: integer
minimum number of pixel in one window to detect the line
nwindows: integer
Number of sliding windows used to search each line
Returns
-------
array
Debug bird eye view image with right and left line marked
array
BGR image with marked lines, debug image in up left corner, radius and car possition printed in up right corner
array
Coeeficients of left line quadratic poly fit
array
Coeeficients of right line quadratic poly fit
"""
#initialize variables used later in the code
offCenter = None
finalRadius =None
leftFit = None
rightFit = None
outputImg = undistImg.copy()
leftFit = [0,0,0]
rightFit = [0,0,0]
# Take a histogram of the bottom half of the image
histogram = np.sum(img[int(img.shape[0]/2):,:], axis=0)
# Create an output image to draw on and visualize the result
debugImg = np.uint8(np.dstack((img, img, img))*255)
# Find the peak of the left and right halves of the histogram
# These will be the starting point for the left and right lines
midpoint = np.int(histogram.shape[0]/2)
leftx_base = np.argmax(histogram[:midpoint])
rightx_base = np.argmax(histogram[midpoint:]) + midpoint
# Set height of windows
window_height = np.int(img.shape[0]/nwindows)
# Identify the x and y positions of all nonzero pixels in the image
nonzero = img.nonzero()
nonzeroy = np.array(nonzero[0])
nonzerox = np.array(nonzero[1])
# Create empty lists to receive left and right lane pixel indices
left_lane_inds = []
right_lane_inds = []
# try to find lines usign previus fit
if prevLeftFit != None and prevRightFit != None:
left_lane_inds = ((nonzerox > calcQuadFitVal(prevLeftFit, nonzeroy) - margin) &
(nonzerox < calcQuadFitVal(prevLeftFit, nonzeroy) + margin))
right_lane_inds = ((nonzerox > calcQuadFitVal(prevRightFit, nonzeroy) - margin) &
(nonzerox < calcQuadFitVal(prevRightFit, nonzeroy) + margin))
lPixX = nonzerox[left_lane_inds]
lPixY = nonzeroy[left_lane_inds]
rPixX = nonzerox[right_lane_inds]
rPixY = nonzeroy[right_lane_inds]
if pixelsSanityCheck(rPixY, minpix, img.shape) and pixelsSanityCheck(lPixY, minpix, img.shape):
leftFit = squarefit(lPixX, lPixY)
rightFit = squarefit(rPixX, rPixY)
if rightFit == None or leftFit == None or lineFitSanitycheck(leftFit, rightFit, margin, img.shape) == False:
left_lane_inds = []
right_lane_inds = []
leftFit = None
rightFit = None
#check if both lines were found where were expected if not execute full search
if len(right_lane_inds) < nwindows*minpix or len(right_lane_inds) < nwindows*minpix:
# Current positions to be updated for each window
leftx_current = leftx_base
rightx_current = rightx_base
# Create empty lists to receive left and right lane pixel indices
left_lane_inds = []
right_lane_inds = []
# Step through the windows one by one
for window in range(nwindows):
# Identify window boundaries in x and y (and right and left)
win_y_low = img.shape[0] - (window+1)*window_height
win_y_high = img.shape[0] - window*window_height
win_xleft_low = leftx_current - margin
win_xleft_high = leftx_current + margin
win_xright_low = rightx_current - margin
win_xright_high = rightx_current + margin
# Draw the windows on the visualization image
# Identify the nonzero pixels in x and y within the window
good_left_inds = ((nonzeroy >= win_y_low) & (nonzeroy < win_y_high) &
(nonzerox >= win_xleft_low) & (nonzerox < win_xleft_high)).nonzero()[0]
good_right_inds = ((nonzeroy >= win_y_low) & (nonzeroy < win_y_high) &
(nonzerox >= win_xright_low) & (nonzerox < win_xright_high)).nonzero()[0]
# If you found > minpix pixels, recenter next window on their mean position
if len(good_left_inds) > minpix:
left_lane_inds.append(good_left_inds)
# Append only when line is detected
leftx_current = np.int(np.mean(nonzerox[good_left_inds]))
cv2.rectangle(debugImg,(win_xleft_low,win_y_low),(win_xleft_high,win_y_high),
(0,255,0), 4)
if len(good_right_inds) > minpix:
right_lane_inds.append(good_right_inds)
# Append only when line is detected
rightx_current = np.int(np.mean(nonzerox[good_right_inds]))
cv2.rectangle(debugImg,(win_xright_low,win_y_low),(win_xright_high,win_y_high),
(0,255,0), 4)
if len(left_lane_inds) !=0 and len(right_lane_inds) !=0 :
# Concatenate the arrays of indices
left_lane_inds = np.concatenate(left_lane_inds)
right_lane_inds = np.concatenate(right_lane_inds)
lPixX = nonzerox[left_lane_inds]
lPixY = nonzeroy[left_lane_inds]
rPixX = nonzerox[right_lane_inds]
rPixY = nonzeroy[right_lane_inds]
if pixelsSanityCheck(rPixY, minpix, img.shape) and pixelsSanityCheck(lPixY, minpix, img.shape):
leftFit = squarefit(lPixX, lPixY)
rightFit = squarefit(rPixX, rPixY)
#if sanity check still fails use line detected in previous frame
if rightFit == None or leftFit == None or lineFitSanitycheck(leftFit, rightFit, margin, img.shape) == False:
leftFit = prevLeftFit
rightFit= prevRightFit
else:
debugImg[nonzeroy[left_lane_inds], nonzerox[left_lane_inds]] = [255, 0, 0]
debugImg[nonzeroy[right_lane_inds], nonzerox[right_lane_inds]] = [0, 0, 255]
if leftFit != None and rightFit != None:
ploty = np.linspace(0, img.shape[0]-1, img.shape[0] )
left_fitx = calcQuadFitVal(leftFit, ploty)
right_fitx = calcQuadFitVal(rightFit, ploty)
cv2.polylines(debugImg, np.int_(np.dstack((left_fitx, ploty))), False, (255,255, 0),5)
cv2.polylines(debugImg, np.int_(np.dstack((right_fitx, ploty))), False, (255,255, 0),5)
Munwrap = cv2.getPerspectiveTransform(dst, src)
#select lines by color
lineMask = cv2.inRange(debugImg, np.array([255,255,0]), np.array([255,255,0]))
img_size = (lineMask.shape[1], lineMask.shape[0])
maskUnwraped = cv2.warpPerspective(lineMask, Munwrap, img_size , flags=cv2.INTER_LINEAR)
outputImg[maskUnwraped.astype(bool)] = np.array([0,0,255]) #overlay markd lines
#Calculate Radius for left and right fit and select one with more pixels
leftRadius = calculateRadius(leftFit, debugImg.shape[0])
rightRadius = calculateRadius(rightFit, debugImg.shape[0])
if len(left_lane_inds) > len(right_lane_inds):
finalRadius = leftRadius
else:
finalRadius = rightRadius
#Calculate how far of center the car is
offCenter = ((leftx_base + rightx_base) - debugImg.shape[1]) * xmPerPix
#print('left', left_fit, 'right', right_fit)
#print(leftRadius, 'm', rightRadius, 'm')
if offCenter != None and finalRadius != None:
cv2.putText(outputImg, 'Radius: ' + str("{0:.1f}".format(finalRadius)) +'m', (670,50), cv2.FONT_HERSHEY_SIMPLEX, 2, (0,0,0), thickness=2)
cv2.putText(outputImg, 'Off Center: ' + str("{0:.1f}".format(offCenter)) +'m', (550,100), cv2.FONT_HERSHEY_SIMPLEX, 2, (0,0,0), thickness=2)
smallDebug = cv2.resize(debugImg, (0, 0), fx=0.4, fy=0.4)
outputImg[20:20 + smallDebug.shape[0], 20:20 + smallDebug.shape[1], :] = smallDebug
return debugImg, outputImg, leftFit, rightFit
prevLeftFit = None # Variable storing previously detected left line
prevRightFit = None # Variable storing previously detected right line
src = np.float32([[580,460], [702,460], [233,700], [1075,700]]) # Source coordinates for bird eye view transformation
offset = 300
img_size = [1280, 720]
dst = np.float32([[offset, 0], [img_size[0]-offset, 0],
[offset, img_size[1]],
[img_size[0]-offset, img_size[1]]])# Destination coordinates for bird eye view transformation
def processImageLines(image):
"""
This functions detect lines, calculate curvature and car position on RGB image
This function combine full pipeline detecting lines on RGB image
Parameters
----------
image : array
Input RGB image to be converted.
Returns
-------
array
Image with detected lines, curvature radius and car possition overlaid.
"""
global prevLeftFit
global prevRightFit
imageBGR = cv2.cvtColor(image, cv2.COLOR_RGB2BGR) #convert image to BGR because image pipeline accepts this format
dist_pickle = pickle.load( open( "camera_cal/wide_dist_pickle.p", "rb" ) ) # read distortion coeficinets calculated above
mtx = dist_pickle["mtx"]
dist = dist_pickle["dist"]
undistImg = cv2.undistort(imageBGR, mtx, dist, None, mtx)
binImg = convertToBinNotGray(undistImg)
img_size = (binImg.shape[1], binImg.shape[0])
M = cv2.getPerspectiveTransform(src, dst)
wrapedImg = cv2.warpPerspective(binImg, M, img_size , flags=cv2.INTER_LINEAR)
markedImg, undistImg, prevLeftFit, prevRightFit = markLines(wrapedImg, undistImg, prevLeftFit = prevLeftFit, prevRightFit = prevRightFit)
return cv2.cvtColor(undistImg, cv2.COLOR_BGR2RGB) #Return RGB image
In [2]:
import numpy as np
import cv2
import matplotlib.image as mpimg
import matplotlib.pyplot as plt
from skimage.feature import hog
from scipy.ndimage.measurements import label
# Define a function to return HOG features and visualization
def get_hog_features(img, orient, pix_per_cell, cell_per_block,
vis=False, feature_vec=True):
# Call with two outputs if vis==True
if vis == True:
features, hog_image = hog(img, orientations=orient,
pixels_per_cell=(pix_per_cell, pix_per_cell),
block_norm= 'L2-Hys',
cells_per_block=(cell_per_block, cell_per_block),
transform_sqrt=True,
visualise=vis, feature_vector=feature_vec)
return features, hog_image
# Otherwise call with one output
else:
features = hog(img, orientations=orient,
pixels_per_cell=(pix_per_cell, pix_per_cell),
cells_per_block=(cell_per_block, cell_per_block),
block_norm= 'L2-Hys',
transform_sqrt=True,
visualise=vis, feature_vector=feature_vec)
return features
# Define a function to compute binned color features
def bin_spatial(img, size=(32, 32)):
# Use cv2.resize().ravel() to create the feature vector
features = cv2.resize(img, size).ravel()
# Return the feature vector
return features
# Define a function to compute color histogram features
# NEED TO CHANGE bins_range if reading .png files with mpimg!
def color_hist(img, nbins=32, bins_range=(0, 256)):
# Compute the histogram of the color channels separately
channel1_hist = np.histogram(img[:,:,0], bins=nbins, range=bins_range)
channel2_hist = np.histogram(img[:,:,1], bins=nbins, range=bins_range)
channel3_hist = np.histogram(img[:,:,2], bins=nbins, range=bins_range)
# Concatenate the histograms into a single feature vector
hist_features = np.concatenate((channel1_hist[0], channel2_hist[0], channel3_hist[0]))
# Return the individual histograms, bin_centers and feature vector
return hist_features
# Define a function to extract features from a single image window
# This function is very similar to extract_features()
# just for a single image rather than list of images
def single_img_features(img, color_space='RGB', spatial_size=(32, 32),
hist_bins=32, orient=9,
pix_per_cell=8, cell_per_block=2, hog_channel=0,
spatial_feat=True, hist_feat=True, hog_feat=True):
#1) Define an empty list to receive features
img_features = []
#2) Apply color conversion if other than 'RGB'
if color_space != 'RGB':
if color_space == 'HSV':
feature_image = cv2.cvtColor(img, cv2.COLOR_RGB2HSV)
elif color_space == 'LUV':
feature_image = cv2.cvtColor(img, cv2.COLOR_RGB2LUV)
elif color_space == 'HLS':
feature_image = cv2.cvtColor(img, cv2.COLOR_RGB2HLS)
elif color_space == 'YUV':
feature_image = cv2.cvtColor(img, cv2.COLOR_RGB2YUV)
elif color_space == 'YCrCb':
feature_image = cv2.cvtColor(img, cv2.COLOR_RGB2YCrCb)
else: feature_image = np.copy(img)
#3) Compute spatial features if flag is set
if spatial_feat == True:
spatial_features = bin_spatial(feature_image, size=spatial_size)
#4) Append features to list
img_features.append(spatial_features)
#5) Compute histogram features if flag is set
if hist_feat == True:
hist_features = color_hist(feature_image, nbins=hist_bins)
#6) Append features to list
img_features.append(hist_features)
#7) Compute HOG features if flag is set
if hog_feat == True:
if hog_channel == 'ALL':
hog_features = []
for channel in range(feature_image.shape[2]):
hog_features.extend(get_hog_features(feature_image[:,:,channel],
orient, pix_per_cell, cell_per_block,
vis=False, feature_vec=True))
else:
hog_features = get_hog_features(feature_image[:,:,hog_channel], orient,
pix_per_cell, cell_per_block, vis=False, feature_vec=True)
#8) Append features to list
img_features.append(hog_features)
#9) Return concatenated array of features
return np.concatenate(img_features)
# Define a function to extract features from a list of images
# Have this function call bin_spatial() and color_hist()
def extract_features(imgs, color_space='RGB', spatial_size=(32, 32),
hist_bins=32, orient=9,
pix_per_cell=8, cell_per_block=2, hog_channel=0,
spatial_feat=True, hist_feat=True, hog_feat=True):
# Create a list to append feature vectors to
features = []
# Iterate through the list of images
for file in imgs:
file_features = []
# Read in each one by one
#print(file)
image = mpimg.imread(file)
feat = single_img_features(image, color_space=color_space, spatial_size=spatial_size,
hist_bins=hist_bins, orient=orient,
pix_per_cell=pix_per_cell, cell_per_block=cell_per_block, hog_channel=hog_channel,
spatial_feat=spatial_feat, hist_feat=hist_feat, hog_feat=hog_feat)
features.append(feat)
# Return list of feature vectors
return features
# Define a function that takes an image,
# start and stop positions in both x and y,
# window size (x and y dimensions),
# and overlap fraction (for both x and y)
def slide_window(img, x_start_stop=[None, None], y_start_stop=[None, None],
xy_windows=[(64, 64)], xy_overlap=(0.5, 0.5)):
# If x and/or y start/stop positions not defined, set to image size
if x_start_stop[0] == None:
x_start_stop[0] = 0
if x_start_stop[1] == None:
x_start_stop[1] = img.shape[1]
if y_start_stop[0] == None:
y_start_stop[0] = 0
if y_start_stop[1] == None:
y_start_stop[1] = img.shape[0]
# Compute the number of windows in x/y
# Initialize a list to append window positions to
window_list = []
# Compute the number of pixels per step in x/y
for xy_window in xy_windows:
x_step = xy_window[0]*(1-xy_overlap[0])
y_step = xy_window[1]*(1-xy_overlap[1])
for y in range (int(y_start_stop[0]), int(y_start_stop[1])-xy_window[1] + 1, int(y_step)):
for x in range(int(x_start_stop[0]), int(x_start_stop[1]) -xy_window[0] + 1, int(x_step)):
window_list.append(((x,y),(x+xy_window[0] ,y+xy_window[1])))
# Return the list of windows
return window_list
# Define a function to draw bounding boxes
def draw_boxes(img, bboxes, color=(0, 0, 255), thick=6):
# Make a copy of the image
imcopy = np.copy(img)
# Iterate through the bounding boxes
for bbox in bboxes:
# Draw a rectangle given bbox coordinates
cv2.rectangle(imcopy, bbox[0], bbox[1], color, thick)
# Return the image copy with boxes drawn
return imcopy
# Define a function you will pass an image
# and the list of windows to be searched (output of slide_windows())
def search_windows(img, windows, clf, scaler, train_img_size=(64,64), color_space='RGB',
spatial_size=(32, 32), hist_bins=32,
hist_range=(0, 256), orient=9,
pix_per_cell=8, cell_per_block=2,
hog_channel=0, spatial_feat=True,
hist_feat=True, hog_feat=True):
#1) Create an empty list to receive positive detection windows
on_windows = []
#2) Iterate over all windows in the list
for window in windows:
#3) Extract the test window from original image
test_img = cv2.resize(img[window[0][1]:window[1][1], window[0][0]:window[1][0]], train_img_size)
#4) Extract features for that window using single_img_features()
features = single_img_features(test_img, color_space=color_space,
spatial_size=spatial_size, hist_bins=hist_bins,
orient=orient, pix_per_cell=pix_per_cell,
cell_per_block=cell_per_block,
hog_channel=hog_channel, spatial_feat=spatial_feat,
hist_feat=hist_feat, hog_feat=hog_feat)
test_features = np.array(features).reshape(1, -1)
#5) scale
#test_features = scaler.transform(test_features)
#6) Predict using your classifier
prediction = clf.predict(test_features)
#7) If positive (prediction == 1) then save the window
if prediction == 1:
on_windows.append(window)
#8) Return windows for positive detections
return on_windows
# Define a single function that can extract features using hog sub-sampling and make predictions
# Define a single function that can extract features using hog sub-sampling and make predictions
def find_cars(img, ystart, ystop, scale, svc, X_scaler, orient, pix_per_cell, cell_per_block,
spatial_size, hist_bins, hog_channel=0, color_space='RGB', spatial_feat=True,
hist_feat=True, hog_feat=True):
hotRectangles = []
img_tosearch = img[ystart:ystop,:,:]
# apply color conversion if other than 'RGB'
if color_space != 'RGB':
if color_space == 'HSV':
ctrans_tosearch = cv2.cvtColor(img_tosearch, cv2.COLOR_RGB2HSV)
elif color_space == 'LUV':
ctrans_tosearch = cv2.cvtColor(img_tosearch, cv2.COLOR_RGB2LUV)
elif color_space == 'HLS':
ctrans_tosearch = cv2.cvtColor(img_tosearch, cv2.COLOR_RGB2HLS)
elif color_space == 'YUV':
ctrans_tosearch = cv2.cvtColor(img_tosearch, cv2.COLOR_RGB2YUV)
elif color_space == 'YCrCb':
ctrans_tosearch = cv2.cvtColor(img_tosearch, cv2.COLOR_RGB2YCrCb)
else: ctrans_tosearch = np.copy(image)
if scale != 1:
imshape = ctrans_tosearch.shape
ctrans_tosearch = cv2.resize(ctrans_tosearch, (np.int(imshape[1]/scale), np.int(imshape[0]/scale)))
# select colorspace channel for HOG
if hog_channel == 'ALL':
ch1 = ctrans_tosearch[:,:,0]
ch2 = ctrans_tosearch[:,:,1]
ch3 = ctrans_tosearch[:,:,2]
else:
ch1 = ctrans_tosearch[:,:,hog_channel]
# Define blocks and steps as above
nxblocks = (ch1.shape[1] // pix_per_cell) - cell_per_block + 1
nyblocks = (ch1.shape[0] // pix_per_cell) - cell_per_block + 1
nfeat_per_block = orient*cell_per_block**2
# 64 was the orginal sampling rate, with 8 cells and 8 pix per cell
window = 64
nblocks_per_window = (window // pix_per_cell) - cell_per_block + 1
cells_per_step = 2 # Instead of overlap, define how many cells to step
nxsteps = (nxblocks - nblocks_per_window) // cells_per_step + 1
nysteps = (nyblocks - nblocks_per_window) // cells_per_step + 1
if hog_feat:
# Compute individual channel HOG features for the entire image
hog1 = get_hog_features(ch1, orient, pix_per_cell, cell_per_block, feature_vec=False)
if hog_channel == 'ALL':
hog2 = get_hog_features(ch2, orient, pix_per_cell, cell_per_block, feature_vec=False)
hog3 = get_hog_features(ch3, orient, pix_per_cell, cell_per_block, feature_vec=False)
for xb in range(nxsteps):
for yb in range(nysteps):
ypos = yb*cells_per_step
xpos = xb*cells_per_step
if hog_feat:
# Extract HOG for this patch
hog_feat1 = hog1[ypos:ypos+nblocks_per_window, xpos:xpos+nblocks_per_window].ravel()
if hog_channel == 'ALL':
hog_feat2 = hog2[ypos:ypos+nblocks_per_window, xpos:xpos+nblocks_per_window].ravel()
hog_feat3 = hog3[ypos:ypos+nblocks_per_window, xpos:xpos+nblocks_per_window].ravel()
hog_features = np.hstack((hog_feat1, hog_feat2, hog_feat3))
else:
hog_features = hog_feat1
else:
hog_features = []
xleft = xpos*pix_per_cell
ytop = ypos*pix_per_cell
# Extract the image patch
subimg = cv2.resize(ctrans_tosearch[ytop:ytop+window, xleft:xleft+window], (64,64))
# Get color features
if spatial_feat:
spatial_features = bin_spatial(subimg, size=spatial_size)
if hist_feat:
hist_features = color_hist(subimg, nbins=hist_bins)
test_features = np.hstack((spatial_features, hist_features, hog_features)).reshape(1, -1)
# Scale features and make a prediction
#test_features = X_scaler.transform(test_features)
test_prediction = svc.predict(test_features)
if test_prediction == 1:
xbox_left = np.int(xleft*scale)
ytop_draw = np.int(ytop*scale)
win_draw = np.int(window*scale)
hotRectangles.append(((xbox_left, ytop_draw+ystart),(xbox_left+win_draw,ytop_draw+win_draw+ystart)))
return hotRectangles
def add_heat(heatmap, bbox_list):
# Iterate through list of bboxes
for box in bbox_list:
# Add += 1 for all pixels inside each bbox
# Assuming each "box" takes the form ((x1, y1), (x2, y2))
heatmap[box[0][1]:box[1][1], box[0][0]:box[1][0]] += 1
# Return updated heatmap
return heatmap# Iterate through list of bboxes
def apply_threshold(heatmap, threshold):
# Zero out pixels below the threshold
heatmap[heatmap <= threshold] = 0
# Return thresholded map
return heatmap
def draw_labeled_bboxes(img, labels):
# Iterate through all detected cars
for car_number in range(1, labels[1]+1):
# Find pixels with each car_number label value
nonzero = (labels[0] == car_number).nonzero()
# Identify x and y values of those pixels
nonzeroy = np.array(nonzero[0])
nonzerox = np.array(nonzero[1])
# Define a bounding box based on min/max x and y
bbox = ((np.min(nonzerox), np.min(nonzeroy)), (np.max(nonzerox), np.max(nonzeroy)))
# Draw the box on the image
cv2.rectangle(img, bbox[0], bbox[1], (0,0,255), 6)
# Return the image
return img
import numpy as np
import cv2
import matplotlib.image as mpimg
import matplotlib.pyplot as plt
from skimage.feature import hog
from scipy.ndimage.measurements import label
# Define a function to return HOG features and visualization
def get_hog_features(img, orient, pix_per_cell, cell_per_block,
vis=False, feature_vec=True):
# Call with two outputs if vis==True
if vis == True:
features, hog_image = hog(img, orientations=orient,
pixels_per_cell=(pix_per_cell, pix_per_cell),
block_norm= 'L2-Hys',
cells_per_block=(cell_per_block, cell_per_block),
transform_sqrt=True,
visualise=vis, feature_vector=feature_vec)
return features, hog_image
# Otherwise call with one output
else:
features = hog(img, orientations=orient,
pixels_per_cell=(pix_per_cell, pix_per_cell),
cells_per_block=(cell_per_block, cell_per_block),
block_norm= 'L2-Hys',
transform_sqrt=True,
visualise=vis, feature_vector=feature_vec)
return features
# Define a function to compute binned color features
def bin_spatial(img, size=(32, 32)):
# Use cv2.resize().ravel() to create the feature vector
features = cv2.resize(img, size).ravel()
# Return the feature vector
return features
# Define a function to compute color histogram features
# NEED TO CHANGE bins_range if reading .png files with mpimg!
def color_hist(img, nbins=32, bins_range=(0, 256)):
# Compute the histogram of the color channels separately
channel1_hist = np.histogram(img[:,:,0], bins=nbins, range=bins_range)
channel2_hist = np.histogram(img[:,:,1], bins=nbins, range=bins_range)
channel3_hist = np.histogram(img[:,:,2], bins=nbins, range=bins_range)
# Concatenate the histograms into a single feature vector
hist_features = np.concatenate((channel1_hist[0], channel2_hist[0], channel3_hist[0]))
# Return the individual histograms, bin_centers and feature vector
return hist_features
# Define a function to extract features from a single image window
# This function is very similar to extract_features()
# just for a single image rather than list of images
def single_img_features(img, color_space='RGB', spatial_size=(32, 32),
hist_bins=32, orient=9,
pix_per_cell=8, cell_per_block=2, hog_channel=0,
spatial_feat=True, hist_feat=True, hog_feat=True):
#1) Define an empty list to receive features
img_features = []
#2) Apply color conversion if other than 'RGB'
if color_space != 'RGB':
if color_space == 'HSV':
feature_image = cv2.cvtColor(img, cv2.COLOR_RGB2HSV)
elif color_space == 'LUV':
feature_image = cv2.cvtColor(img, cv2.COLOR_RGB2LUV)
elif color_space == 'HLS':
feature_image = cv2.cvtColor(img, cv2.COLOR_RGB2HLS)
elif color_space == 'YUV':
feature_image = cv2.cvtColor(img, cv2.COLOR_RGB2YUV)
elif color_space == 'YCrCb':
feature_image = cv2.cvtColor(img, cv2.COLOR_RGB2YCrCb)
else: feature_image = np.copy(img)
#3) Compute spatial features if flag is set
if spatial_feat == True:
spatial_features = bin_spatial(feature_image, size=spatial_size)
#4) Append features to list
img_features.append(spatial_features)
#5) Compute histogram features if flag is set
if hist_feat == True:
hist_features = color_hist(feature_image, nbins=hist_bins)
#6) Append features to list
img_features.append(hist_features)
#7) Compute HOG features if flag is set
if hog_feat == True:
if hog_channel == 'ALL':
hog_features = []
for channel in range(feature_image.shape[2]):
hog_features.extend(get_hog_features(feature_image[:,:,channel],
orient, pix_per_cell, cell_per_block,
vis=False, feature_vec=True))
else:
hog_features = get_hog_features(feature_image[:,:,hog_channel], orient,
pix_per_cell, cell_per_block, vis=False, feature_vec=True)
#8) Append features to list
img_features.append(hog_features)
#9) Return concatenated array of features
return np.concatenate(img_features)
# Define a function to extract features from a list of images
# Have this function call bin_spatial() and color_hist()
def extract_features(imgs, color_space='RGB', spatial_size=(32, 32),
hist_bins=32, orient=9,
pix_per_cell=8, cell_per_block=2, hog_channel=0,
spatial_feat=True, hist_feat=True, hog_feat=True):
# Create a list to append feature vectors to
features = []
# Iterate through the list of images
for file in imgs:
file_features = []
# Read in each one by one
#print(file)
image = mpimg.imread(file)
feat = single_img_features(image, color_space=color_space, spatial_size=spatial_size,
hist_bins=hist_bins, orient=orient,
pix_per_cell=pix_per_cell, cell_per_block=cell_per_block, hog_channel=hog_channel,
spatial_feat=spatial_feat, hist_feat=hist_feat, hog_feat=hog_feat)
features.append(feat)
# Return list of feature vectors
return features
# Define a function that takes an image,
# start and stop positions in both x and y,
# window size (x and y dimensions),
# and overlap fraction (for both x and y)
def slide_window(img, x_start_stop=[None, None], y_start_stop=[None, None],
xy_windows=[(64, 64)], xy_overlap=(0.5, 0.5)):
# If x and/or y start/stop positions not defined, set to image size
if x_start_stop[0] == None:
x_start_stop[0] = 0
if x_start_stop[1] == None:
x_start_stop[1] = img.shape[1]
if y_start_stop[0] == None:
y_start_stop[0] = 0
if y_start_stop[1] == None:
y_start_stop[1] = img.shape[0]
# Compute the number of windows in x/y
# Initialize a list to append window positions to
window_list = []
# Compute the number of pixels per step in x/y
for xy_window in xy_windows:
x_step = xy_window[0]*(1-xy_overlap[0])
y_step = xy_window[1]*(1-xy_overlap[1])
for y in range (int(y_start_stop[0]), int(y_start_stop[1])-xy_window[1] + 1, int(y_step)):
for x in range(int(x_start_stop[0]), int(x_start_stop[1]) -xy_window[0] + 1, int(x_step)):
window_list.append(((x,y),(x+xy_window[0] ,y+xy_window[1])))
# Return the list of windows
return window_list
# Define a function to draw bounding boxes
def draw_boxes(img, bboxes, color=(0, 0, 255), thick=6):
# Make a copy of the image
imcopy = np.copy(img)
# Iterate through the bounding boxes
for bbox in bboxes:
# Draw a rectangle given bbox coordinates
cv2.rectangle(imcopy, bbox[0], bbox[1], color, thick)
# Return the image copy with boxes drawn
return imcopy
# Define a function you will pass an image
# and the list of windows to be searched (output of slide_windows())
def search_windows(img, windows, clf, scaler, train_img_size=(64,64), color_space='RGB',
spatial_size=(32, 32), hist_bins=32,
hist_range=(0, 256), orient=9,
pix_per_cell=8, cell_per_block=2,
hog_channel=0, spatial_feat=True,
hist_feat=True, hog_feat=True):
#1) Create an empty list to receive positive detection windows
on_windows = []
#2) Iterate over all windows in the list
for window in windows:
#3) Extract the test window from original image
test_img = cv2.resize(img[window[0][1]:window[1][1], window[0][0]:window[1][0]], train_img_size)
#4) Extract features for that window using single_img_features()
features = single_img_features(test_img, color_space=color_space,
spatial_size=spatial_size, hist_bins=hist_bins,
orient=orient, pix_per_cell=pix_per_cell,
cell_per_block=cell_per_block,
hog_channel=hog_channel, spatial_feat=spatial_feat,
hist_feat=hist_feat, hog_feat=hog_feat)
test_features = np.array(features).reshape(1, -1)
#5) scale
#test_features = scaler.transform(test_features)
#6) Predict using your classifier
prediction = clf.predict(test_features)
#7) If positive (prediction == 1) then save the window
if prediction == 1:
on_windows.append(window)
#8) Return windows for positive detections
return on_windows
# Define a single function that can extract features using hog sub-sampling and make predictions
# Define a single function that can extract features using hog sub-sampling and make predictions
def find_cars(img, ystart, ystop, scale, svc, X_scaler, orient, pix_per_cell, cell_per_block,
spatial_size, hist_bins, hog_channel=0, color_space='RGB', spatial_feat=True,
hist_feat=True, hog_feat=True):
hotRectangles = []
img_tosearch = img[ystart:ystop,:,:]
# apply color conversion if other than 'RGB'
if color_space != 'RGB':
if color_space == 'HSV':
ctrans_tosearch = cv2.cvtColor(img_tosearch, cv2.COLOR_RGB2HSV)
elif color_space == 'LUV':
ctrans_tosearch = cv2.cvtColor(img_tosearch, cv2.COLOR_RGB2LUV)
elif color_space == 'HLS':
ctrans_tosearch = cv2.cvtColor(img_tosearch, cv2.COLOR_RGB2HLS)
elif color_space == 'YUV':
ctrans_tosearch = cv2.cvtColor(img_tosearch, cv2.COLOR_RGB2YUV)
elif color_space == 'YCrCb':
ctrans_tosearch = cv2.cvtColor(img_tosearch, cv2.COLOR_RGB2YCrCb)
else: ctrans_tosearch = np.copy(image)
if scale != 1:
imshape = ctrans_tosearch.shape
ctrans_tosearch = cv2.resize(ctrans_tosearch, (np.int(imshape[1]/scale), np.int(imshape[0]/scale)))
# select colorspace channel for HOG
if hog_channel == 'ALL':
ch1 = ctrans_tosearch[:,:,0]
ch2 = ctrans_tosearch[:,:,1]
ch3 = ctrans_tosearch[:,:,2]
else:
ch1 = ctrans_tosearch[:,:,hog_channel]
# Define blocks and steps as above
nxblocks = (ch1.shape[1] // pix_per_cell) - cell_per_block + 1
nyblocks = (ch1.shape[0] // pix_per_cell) - cell_per_block + 1
nfeat_per_block = orient*cell_per_block**2
# 64 was the orginal sampling rate, with 8 cells and 8 pix per cell
window = 64
nblocks_per_window = (window // pix_per_cell) - cell_per_block + 1
cells_per_step = 2 # Instead of overlap, define how many cells to step
nxsteps = (nxblocks - nblocks_per_window) // cells_per_step + 1
nysteps = (nyblocks - nblocks_per_window) // cells_per_step + 1
if hog_feat:
# Compute individual channel HOG features for the entire image
hog1 = get_hog_features(ch1, orient, pix_per_cell, cell_per_block, feature_vec=False)
if hog_channel == 'ALL':
hog2 = get_hog_features(ch2, orient, pix_per_cell, cell_per_block, feature_vec=False)
hog3 = get_hog_features(ch3, orient, pix_per_cell, cell_per_block, feature_vec=False)
for xb in range(nxsteps):
for yb in range(nysteps):
ypos = yb*cells_per_step
xpos = xb*cells_per_step
if hog_feat:
# Extract HOG for this patch
hog_feat1 = hog1[ypos:ypos+nblocks_per_window, xpos:xpos+nblocks_per_window].ravel()
if hog_channel == 'ALL':
hog_feat2 = hog2[ypos:ypos+nblocks_per_window, xpos:xpos+nblocks_per_window].ravel()
hog_feat3 = hog3[ypos:ypos+nblocks_per_window, xpos:xpos+nblocks_per_window].ravel()
hog_features = np.hstack((hog_feat1, hog_feat2, hog_feat3))
else:
hog_features = hog_feat1
else:
hog_features = []
xleft = xpos*pix_per_cell
ytop = ypos*pix_per_cell
# Extract the image patch
subimg = cv2.resize(ctrans_tosearch[ytop:ytop+window, xleft:xleft+window], (64,64))
# Get color features
if spatial_feat:
spatial_features = bin_spatial(subimg, size=spatial_size)
if hist_feat:
hist_features = color_hist(subimg, nbins=hist_bins)
test_features = np.hstack((spatial_features, hist_features, hog_features)).reshape(1, -1)
# Scale features and make a prediction
#test_features = X_scaler.transform(test_features)
test_prediction = svc.predict(test_features)
if test_prediction == 1:
xbox_left = np.int(xleft*scale)
ytop_draw = np.int(ytop*scale)
win_draw = np.int(window*scale)
hotRectangles.append(((xbox_left, ytop_draw+ystart),(xbox_left+win_draw,ytop_draw+win_draw+ystart)))
return hotRectangles
def add_heat(heatmap, bbox_list):
# Iterate through list of bboxes
for box in bbox_list:
# Add += 1 for all pixels inside each bbox
# Assuming each "box" takes the form ((x1, y1), (x2, y2))
heatmap[box[0][1]:box[1][1], box[0][0]:box[1][0]] += 1
# Return updated heatmap
return heatmap# Iterate through list of bboxes
def apply_threshold(heatmap, threshold):
# Zero out pixels below the threshold
heatmap[heatmap <= threshold] = 0
# Return thresholded map
return heatmap
def drawLabeledBboxesIfBoth(img, labels, ifLabel):
# Iterate through all detected cars
for car_number in range(1, labels[1]+1):
# Find pixels with each car_number label value
nonzero = (labels[0] == car_number).nonzero()
# Identify x and y values of those pixels
if any((ifLabel[0][nonzero]!=0)):
nonzeroy = np.array(nonzero[0])
nonzerox = np.array(nonzero[1])
# Define a bounding box based on min/max x and y
bbox = ((np.min(nonzerox), np.min(nonzeroy)), (np.max(nonzerox), np.max(nonzeroy)))
# Draw the box on the image
cv2.rectangle(img, bbox[0], bbox[1], (0,0,255), 6)
# Return the image
return img
Find the best Hog parameters¶
In [ ]:
import matplotlib.image as mpimg
import matplotlib.pyplot as plt
import numpy as np
import cv2
import glob
import time
from sklearn.svm import LinearSVC
from sklearn.svm import SVC
from sklearn.model_selection import GridSearchCV
from sklearn.preprocessing import StandardScaler
from skimage.feature import hog
import pickle
# NOTE: the next import is only valid for scikit-learn version <= 0.17
# for scikit-learn >= 0.18 use:
from sklearn.model_selection import train_test_split
# Read in cars and notcars
notcars = glob.glob('non-vehicles/**/*.png', recursive=True)
cars = glob.glob('vehicles/**/*.png', recursive=True)
print('Number of vehicles:', len(cars), 'Number of non vehicles:', len(notcars))
color_space = 'LUV' # Can be RGB, HSV, LUV, HLS, YUV, YCrCb
orientList = [6,8,10] # HOG orientations
pix_per_cellList = [8,10,16] # HOG pixels per cell
cell_per_blockList = [1,2,3] # HOG cells per block
hog_channel = 0 # Can be 0, 1, 2, or "ALL"
spatial_size = (16, 16) # Spatial binning dimensions
hist_bins = 32 # Number of histogram bins
spatial_feat = False # Spatial features on or off
hist_feat = False # Histogram features on or off
hog_feat = True # HOG features on or off
cParamList = [1,3,6,10,20]
orientBest = 0 # best orientations
pix_per_cellBest = 0 # best pixels per cell
cell_per_blockBest = 0 # best cells per block
hog_channelBest = 0 # best
cParamBest = 0 # best C parameter
accuracyBest = 0
svcBest = 0
# Split up data into randomized training and test sets always the same so accuracy can be compared
rand_state = np.random.randint(0, 100)
for orient in orientList:
for pix_per_cell in pix_per_cellList:
for cell_per_block in cell_per_blockList:
car_features = extract_features(cars, color_space=color_space,
spatial_size=spatial_size, hist_bins=hist_bins,
orient=orient, pix_per_cell=pix_per_cell,
cell_per_block=cell_per_block,
hog_channel=hog_channel, spatial_feat=spatial_feat,
hist_feat=hist_feat, hog_feat=hog_feat)
notcar_features = extract_features(notcars, color_space=color_space,
spatial_size=spatial_size, hist_bins=hist_bins,
orient=orient, pix_per_cell=pix_per_cell,
cell_per_block=cell_per_block,
hog_channel=hog_channel, spatial_feat=spatial_feat,
hist_feat=hist_feat, hog_feat=hog_feat)
# Create an array stack of feature vectors
X = np.vstack((car_features, notcar_features)).astype(np.float64)
# Define the labels vector
y = np.hstack((np.ones(len(car_features)), np.zeros(len(notcar_features))))
X_train, X_test, y_train, y_test = train_test_split(
X, y, test_size=0.2, random_state=rand_state)
# Fit a per-column scaler
X_scaler_train = StandardScaler().fit(X_train)
# Apply the scaler to X
#X_train = X_scaler_train.transform(X_train)
#X_test = X_scaler_train.transform(X_test)
print('Using:',orient,'orientations',pix_per_cell,
'pixels per cell and', cell_per_block,'cells per block')
print('Feature vector length:', len(X_train[0]))
# Use a linear SVC
#parameters = {'kernel':('linear', 'rbf'), 'C':[1, 10]}
#svr = SVC()
#svc = GridSearchCV(svr, parameters)
for cParam in cParamList:
svc = LinearSVC(C=cParam)
# Check the training time for the SVC
t=time.time()
svc.fit(X_train, y_train)
t2 = time.time()
#print('best SVC params', svc.best_params_)
print(round(t2-t, 2), 'Seconds to train SVC...')
# Check the score of the SVC
accuracy = round(svc.score(X_test, y_test), 4)
print('Test Accuracy of SVC = ', accuracy)
if accuracyBest < accuracy:
orientBest = orient # best orientations so far
pix_per_cellBest = pix_per_cell # best pixels per cell so far
cell_per_blockBest = cell_per_block # best cells per block so far
hog_channelBest = hog_channel # best so far
cParamBest = cParam # best so far
accuracyBest = accuracy # store best accuracy
svcBest = svc
# Check the prediction time for a single sample
t=time.time()
print('Best orient:', orientBest)
print('Best pix_per_cell:', pix_per_cellBest)
print('Best cell_per_block:', cell_per_blockBest)
print('Best hog_channel:', hog_channelBest)
print('Best C param:', cParamBest)
# Save the best parameters
dist_pickle = {}
dist_pickle["orient"] = orientBest
dist_pickle["pix_per_cell"] = pix_per_cellBest
dist_pickle["cell_per_block"] = cell_per_blockBest
dist_pickle["Cparam"] = cParamBest
dist_pickle["model"] = svcBest
pickle.dump( dist_pickle, open( "theBestHogParam.p", "wb" ) )
Example hog features¶
In [ ]:
exampleCar = 'vehicles/GTI_Far/image0005.png'
exampleNonCar = 'non-vehicles/Extras/extra6.png'
imageCar = mpimg.imread(exampleCar)
nonImageCar = mpimg.imread(exampleNonCar)
imageCarConv = cv2.cvtColor(imageCar, cv2.COLOR_RGB2LUV)
nonImageCarConv = cv2.cvtColor(nonImageCar, cv2.COLOR_RGB2LUV)
orient = 10 # HOG orientations
pix_per_cell = 10 # HOG pixels per cell
cell_per_block = 3 # HOG cells per block
hog_channel = 0 # Can be 0, 1, 2, or "ALL"
features, carHog = get_hog_features(imageCarConv[:,:,hog_channel], orient, pix_per_cell, cell_per_block,
vis=True, feature_vec=True)
features, nonCarHog = get_hog_features(nonImageCarConv[:,:,hog_channel], orient, pix_per_cell, cell_per_block,
vis=True, feature_vec=True)
f, (ax1, ax2) = plt.subplots(1, 2, figsize=(20,10))
ax1.imshow(imageCar)
ax1.set_title('Car', fontsize=30)
ax2.imshow(carHog, cmap='hot')
ax2.set_title('Car Hog', fontsize=30)
f, (ax1, ax2) = plt.subplots(1, 2, figsize=(20,10))
ax1.imshow(nonImageCar)
ax1.set_title('Non Car', fontsize=30)
ax2.imshow(nonCarHog, cmap='hot')
ax2.set_title('Non Car Hog', fontsize=30)
Train the classifier¶
In [4]:
import matplotlib.image as mpimg
import matplotlib.pyplot as plt
import numpy as np
import cv2
import glob
import time
from sklearn.svm import LinearSVC
from sklearn.svm import SVC
from sklearn.model_selection import GridSearchCV
from sklearn.preprocessing import StandardScaler
from skimage.feature import hog
# NOTE: the next import is only valid for scikit-learn version <= 0.17
# for scikit-learn >= 0.18 use:
from sklearn.model_selection import train_test_split
# Read in cars and notcars
notcars = glob.glob('non-vehicles/**/*.png', recursive=True)
cars = glob.glob('vehicles/**/*.png', recursive=True)
print('Number of vehicles:', len(cars), 'Number of non vehicles:', len(notcars))
dist_pickle = pickle.load( open( "theBestHogParam.p", "rb" ) ) # read distortion coeficinets calculated above
orient = dist_pickle["orient"] # Best HOG orientations
pix_per_cell = dist_pickle["pix_per_cell"] # Best HOG pixels per cell
cell_per_block = dist_pickle["cell_per_block"] # Best HOG cells per block
cParam = dist_pickle["Cparam"] # bestCParam
color_space = 'LUV' # Can be RGB, HSV, LUV, HLS, YUV, YCrCb
hog_channel = 0 # Can be 0, 1, 2, or "ALL"
spatial_size = (16, 16) # Spatial binning dimensions
hist_bins = 32 # Number of histogram bins
spatial_feat = True # Spatial features on or off
hist_feat = True # Histogram features on or off
hog_feat = True # HOG features on or off
car_features = extract_features(cars, color_space=color_space,
spatial_size=spatial_size, hist_bins=hist_bins,
orient=orient, pix_per_cell=pix_per_cell,
cell_per_block=cell_per_block,
hog_channel=hog_channel, spatial_feat=spatial_feat,
hist_feat=hist_feat, hog_feat=hog_feat)
notcar_features = extract_features(notcars, color_space=color_space,
spatial_size=spatial_size, hist_bins=hist_bins,
orient=orient, pix_per_cell=pix_per_cell,
cell_per_block=cell_per_block,
hog_channel=hog_channel, spatial_feat=spatial_feat,
hist_feat=hist_feat, hog_feat=hog_feat)
# Create an array stack of feature vectors
X = np.vstack((car_features, notcar_features)).astype(np.float64)
# Define the labels vector
y = np.hstack((np.ones(len(car_features)), np.zeros(len(notcar_features))))
# Split up data into randomized training and test sets
rand_state = np.random.randint(0, 100)
X_train, X_test, y_train, y_test = train_test_split(
X, y, test_size=0.2, random_state=rand_state)
# Fit a per-column scaler
X_scaler_train = StandardScaler().fit(X_train)
# Apply the scaler to X
#X_train = X_scaler_train.transform(X_train)
#X_test = X_scaler_train.transform(X_test)
print('Using:',orient,'orientations',pix_per_cell,
'pixels per cell and', cell_per_block,'cells per block')
print('Feature vector length:', len(X_train[0]))
# Use a linear SVC
#parameters = {'kernel':('linear', 'rbf'), 'C':[1, 10]}
#svr = SVC()
#svc = GridSearchCV(svr, parameters)
svc = LinearSVC(C=cParam)
# Check the training time for the SVC
t=time.time()
svc.fit(X_train, y_train)
t2 = time.time()
#print('best SVC params', svc.best_params_)
print(round(t2-t, 2), 'Seconds to train SVC...')
# Check the score of the SVC
print('Test Accuracy of SVC = ', round(svc.score(X_test, y_test), 4))
# Check the prediction time for a single sample
t=time.time()
modelPickle["model"] = svc
pickle.dump( modelPickle, open( "trainedModel.p", "wb" ) )
Undistort input pictures¶
In [11]:
import pickle
dist_pickle = pickle.load( open( "camera_cal/wide_dist_pickle.p", "rb" ) ) # read distortion coeficinets calculated above
mtx = dist_pickle["mtx"]
dist = dist_pickle["dist"]
images = glob.glob('test_images/*.jpg')
undist = []
#load trained model
modelPickle = pickle.load( open( "trainedModel.p", "rb" ) )
svc = modelPickle["model"]
for imgName in images:
img = mpimg.imread(imgName)
undistImg = cv2.undistort(img, mtx, dist, None, mtx)
undistImg = undistImg.astype(np.float32)/255# because input is jpg and train was png
undist.append(undistImg)
Test clasifier with fixed window size¶
In [12]:
for image in undist:
draw_image = np.copy(image)
y_start_stop = [400, 680] # Min and max in y to search in slide_window()
windows = slide_window(image, x_start_stop=[None, None], y_start_stop=y_start_stop,
xy_windows=[(100, 100)], xy_overlap=(0.5, 0.5))
hot_windows = search_windows(image, windows, svc, scaler=X_scaler_train, train_img_size=(64,64), color_space=color_space,
spatial_size=spatial_size, hist_bins=hist_bins,
orient=orient, pix_per_cell=pix_per_cell,
cell_per_block=cell_per_block,
hog_channel=hog_channel, spatial_feat=spatial_feat,
hist_feat=hist_feat, hog_feat=hog_feat)
window_img = draw_boxes(draw_image, hot_windows, color=(0, 0, 255), thick=6)
plt.figure()
plt.imshow(window_img)
Test clasifier with variable window size¶
In [13]:
hotWindowsList = []
markedImgsList =[]
for image in undist:
draw_image = np.copy(image)
hot_windows = find_cars(image, 390, 490, 1.2, svc, X_scaler_train, orient, pix_per_cell, cell_per_block, spatial_size,
hist_bins, hog_channel=hog_channel, color_space=color_space, spatial_feat=spatial_feat,
hist_feat=hist_feat, hog_feat=hog_feat)
hot_windows += find_cars(image, 450, 590, 1.4, svc, X_scaler_train, orient, pix_per_cell, cell_per_block, spatial_size,
hist_bins, hog_channel=hog_channel, color_space=color_space, spatial_feat=spatial_feat,
hist_feat=hist_feat, hog_feat=hog_feat)
hot_windows += find_cars(image, 560, 680, 1.7, svc, X_scaler_train, orient, pix_per_cell, cell_per_block, spatial_size,
hist_bins, hog_channel=hog_channel, color_space=color_space, spatial_feat=spatial_feat,
hist_feat=hist_feat, hog_feat=hog_feat)
hotWindowsList.append(hot_windows)
window_img = draw_boxes(draw_image, hot_windows, color=(0, 0, 255), thick=6)
markedImgsList.append(window_img)
plt.figure()
plt.imshow(window_img)
Create heat map¶
In [14]:
for image, hotWindows in zip(undist, hotWindowsList):
heat = np.zeros_like(image[:,:,0]).astype(np.float)
# Add heat to each box in box list
heat = add_heat(heat, hotWindows)
# Apply threshold to help remove false positives
heat = apply_threshold(heat, 2)
# Visualize the heatmap when displaying
heatmap = np.clip(heat, 0, 255)
# Find final boxes from heatmap using label function
labels = label(heatmap)
draw_img = draw_labeled_bboxes(np.copy(image), labels)
f, (ax1, ax2) = plt.subplots(1, 2, figsize=(20,10))
ax1.imshow(draw_img)
ax1.set_title('Marked Img', fontsize=30)
ax2.imshow(heatmap, cmap='hot')
ax2.set_title('Heat Map', fontsize=30)
Video processing function¶
In [15]:
def processImageVehicles(image):
"""
This functions detect vehicles,on RGB image
This function combine full pipeline detecting vehicles on RGB image
Parameters
----------
image : array
Input RGB image to be converted.
Returns
-------
array
Image with marked vehicles
"""
global prevHeat
imageLines = processImageLines(image)
image = image.astype(np.float32)/255# because input is jpg and train was png
hot_windows = find_cars(image, 390, 490, 1.2, svc, X_scaler_train, orient, pix_per_cell, cell_per_block, spatial_size,
hist_bins, hog_channel=hog_channel, color_space=color_space, spatial_feat=spatial_feat,
hist_feat=hist_feat, hog_feat=hog_feat)
hot_windows += find_cars(image, 450, 590, 1.4, svc, X_scaler_train, orient, pix_per_cell, cell_per_block, spatial_size,
hist_bins, hog_channel=hog_channel, color_space=color_space, spatial_feat=spatial_feat,
hist_feat=hist_feat, hog_feat=hog_feat)
hot_windows += find_cars(image, 560, 680, 1.7, svc, X_scaler_train, orient, pix_per_cell, cell_per_block, spatial_size,
hist_bins, hog_channel=hog_channel, color_space=color_space, spatial_feat=spatial_feat,
hist_feat=hist_feat, hog_feat=hog_feat)
heat = np.zeros_like(image[:,:,0]).astype(np.float)
# Add heat to each box in box list
heat = add_heat(heat, hot_windows)
prevHeat.append(heat)
heatCombined = sum(prevHeat)
# Apply threshold to help remove false positives from combined heat
heatCombined = apply_threshold(heatCombined, int(len(prevHeat)*3.3))
heatmapCombined = np.clip(heatCombined, 0, 255)
# Find possible positions on combined map
labelsCombined = label(heatmapCombined)
# Apply threshold to help remove false positives
heat = apply_threshold(heat, 2)
heatmap = np.clip(heat, 0, 255)
# Find final boxes from heatmap using label function
labels = label(heatmap)
draw_img = drawLabeledBboxesIfBoth(np.copy(imageLines), labels, labelsCombined)
return draw_img
Video processing¶
In [16]:
from moviepy.editor import VideoFileClip
from IPython.display import HTML
import collections
%matplotlib qt
%matplotlib inline
prevHeat = collections.deque(maxlen=40) # avreage 10 frames to get rid of false positives
videoName = 'project_video.mp4'
finalVidOutput = 'output_videos/final_projct_video.mp4'
clip1 = VideoFileClip(videoName)
prevLeftFit = None
prevRightFit = None
whiteClip = clip1.fl_image(processImageVehicles)
%time whiteClip.write_videofile(finalVidOutput, audio=False)
Show final video¶
In [17]:
HTML("""
<video width="960" height="540" controls>
<source src="{0}">
</video>
""".format(finalVidOutput))
Out[17]: